Quick Start

Get up and running with Game Reasoning Arena in minutes.

Installation

git clone https://github.com/SLAMPAI/game_reasoning_arena.git
cd game_reasoning_arena
conda env create -f environment.yaml
conda activate game_reasoning_arena
pip install -e .

Your First Game

Run a simple game using the command-line interface:

# Run a Tic-Tac-Toe game with random agents
python scripts/runner.py --config src/game_reasoning_arena/configs/example_config.yaml --override \
  env_configs.0.game_name=tic_tac_toe \
  agents.player_0.type=random \
  agents.player_1.type=random \
  num_episodes=1

# Run a Connect Four game
python scripts/runner.py --config src/game_reasoning_arena/configs/example_config.yaml --override \
  env_configs.0.game_name=connect_four \
  agents.player_0.type=random \
  agents.player_1.type=random \
  num_episodes=1

LLM vs Random Agent

Try an LLM agent against a random player:

python scripts/runner.py --config src/game_reasoning_arena/configs/example_config.yaml --override \
  env_configs.0.game_name=kuhn_poker \
  agents.player_0.type=llm \
  agents.player_0.model=litellm_groq/llama3-8b-8192 \
  agents.player_1.type=random \
  num_episodes=5

Parallel Execution with Ray

For faster experiments, enable Ray parallelization:

# Single model with parallel episodes and games
python scripts/runner.py --config src/game_reasoning_arena/configs/human_vs_random_config.yaml \
  --override use_ray=true parallel_episodes=true

# Multiple models in parallel (maximum speed)
python scripts/run_ray_multi_model.py \
  --config src/game_reasoning_arena/configs/ray_multi_model.yaml \
  --override use_ray=true

What’s Next?